侧面矫正，仿射变换

• This example program shows how to rectify a part of an image using
• vector_to_proj_hom_mat2d. This is done by extracting the four
• corners of a soccer field from an oblique view of a stadium. The four
• corners define a quadrilateral, which is then mapped to a rectangle
• that has the correct aspect ratio. If the dimensions of the field were
• known metrically, the program could be extended to perform metric
• measurements on the soccer field and stadium track.
  dev_update_off ()
  read_image (Image, ‘olympic_stadium’)
  get_image_size (Image, Width, Height)
  dev_close_window ()
  dev_open_window (0, 0, Width, Height, ‘black’, WindowHandle)
  set_display_font (WindowHandle, 16, ‘mono’, ‘true’, ‘false’)
  dev_display (Image)
  disp_message (WindowHandle, ‘Oblique view of stadium track and field’, ‘image’, -1, -1, ‘black’, ‘true’)
  disp_continue_message (WindowHandle, ‘black’, ‘true’)
  stop ()
• Segment the field based on its green color and its shape. This is done
• most easily by using the Hue channel of the HSV transform of the image.
  decompose3 (Image, ImageR, ImageG, ImageB)
  trans_from_rgb (ImageR, ImageG, ImageB, ImageH, ImageS, ImageV, ‘hsv’)
  threshold (ImageH, GreenRegion, 60, 90)
  connection (GreenRegion, ConnectedRegions)
• Select the field region based on the fact that it is the largest component
• that roughly looks like a rotated rectangle.
  select_shape_std (ConnectedRegions, RectangleRegions, ‘rectangle2’, 80)
  select_shape_std (RectangleRegions, Field, ‘max_area’, 90)
• Construct a ROI for the color edge extraction from the boundary of the
• segmented field.
  shape_trans (Field, FieldConvex, ‘convex’)
  boundary (FieldConvex, FieldBorder, ‘inner’)
  dilation_rectangle1 (FieldBorder, FieldDilation, 7, 7)
  dev_display (Image)
  dev_set_draw (‘margin’)
  dev_display (FieldDilation)
  dev_set_draw (‘fill’)
  disp_message (WindowHandle, ‘Field ROI based on color segmentation’, ‘image’, -1, -1, ‘black’, ‘true’)
  disp_continue_message (WindowHandle, ‘black’, ‘true’)
  stop ()
• Perform the color edge extraction.
  reduce_domain (Image, FieldDilation, ImageReduced)
  edges_color_sub_pix (ImageReduced, Edges, ‘canny’, 1.5, 20, 40)
  dev_display (Image)
  dev_display (Edges)
  disp_message (WindowHandle, ‘Extracted color edges’, ‘image’, -1, -1, ‘black’, ‘true’)
  disp_continue_message (WindowHandle, ‘black’, ‘true’)
  stop ()
• Now select straight edges from the extracted edges and group them into
• the four sides of the quadrilateral.
  select_shape_xld (Edges, SelectedEdges, ‘contlength’, ‘and’, 15, 500)
• We first need to segment the edges into straight lines because the
• extracted edges may represent more than one side of the quadrilateral.
  segment_contours_xld (SelectedEdges, ContoursSplit, ‘lines’, 5, 4, 2)
  regress_contours_xld (ContoursSplit, RegressContours, ‘no’, 1)
• Weed out edges that are not straight enough.
  select_contours_xld (RegressContours, SelectedContours, ‘curvature’, 0, 0.5, 0, 0)
• Select edges that are long enough to provide meaningful direction information.
  select_shape_xld (SelectedContours, SelectedEdges, ‘contlength’, ‘and’, 15, 500)
• Group the straight lines into the quadrilateral’s sides.
  union_collinear_contours_xld (SelectedEdges, UnionContours, 30, 1, 4, 0.1, ‘attr_forget’)
  dev_display (Image)
  dev_display (UnionContours)
  disp_message (WindowHandle, ‘Selected straight edges’, ‘image’, -1, -1, ‘black’, ‘true’)
  disp_continue_message (WindowHandle, ‘black’, ‘true’)
  stop ()
• Now we need to determine the order of the sides because we want to
• intersect the sides to get the corners of the quadrilateral. First, we
• separate the sides into roughly horizontal and roughly vertical sides.
  select_shape_xld (UnionContours, HorizontalEdges, ‘phi’, ‘and’, rad(-20), rad(20))
  select_shape_xld (UnionContours, VerticalEdges, [‘phi’,‘phi’], ‘or’, [rad(-90),rad(60)], [rad(-60),rad(90)])
• Now we need to order the sides. The horizontal sides are sorted based on
• their row coordinates, while the vertical sides are sorted based on their
• column coordinates.
  fit_line_contour_xld (HorizontalEdges, ‘tukey’, -1, 0, 10, 2, RowBegHor, ColBegHor, RowEndHor, ColEndHor, NrHor, NcHor, DistHor)
  IndexHor := sort_index(RowBegHor)
  fit_line_contour_xld (VerticalEdges, ‘tukey’, -1, 0, 10, 2, RowBegVer, ColBegVer, RowEndVer, ColEndVer, NrVer, NcVer, DistVer)
  IndexVer := sort_index(ColBegVer)
• After we have sorted the sides, we can intersect the appropriate sides
• to get the four corners of the quadrilateral.
  intersection_lines (RowBegHor[IndexHor[0]], ColBegHor[IndexHor[0]], RowEndHor[IndexHor[0]], ColEndHor[IndexHor[0]], RowBegVer[IndexVer[0]], ColBegVer[IndexVer[0]], RowEndVer[IndexVer[0]], ColEndVer[IndexVer[0]], RowUL, ColUL, IsOverlapping)
  intersection_lines (RowBegHor[IndexHor[0]], ColBegHor[IndexHor[0]], RowEndHor[IndexHor[0]], ColEndHor[IndexHor[0]], RowBegVer[IndexVer[1]], ColBegVer[IndexVer[1]], RowEndVer[IndexVer[1]], ColEndVer[IndexVer[1]], RowUR, ColUR, IsOverlapping)
  intersection_lines (RowBegHor[IndexHor[1]], ColBegHor[IndexHor[1]], RowEndHor[IndexHor[1]], ColEndHor[IndexHor[1]], RowBegVer[IndexVer[0]], ColBegVer[IndexVer[0]], RowEndVer[IndexVer[0]], ColEndVer[IndexVer[0]], RowLL, ColLL, IsOverlapping)
  intersection_lines (RowBegHor[IndexHor[1]], ColBegHor[IndexHor[1]], RowEndHor[IndexHor[1]], ColEndHor[IndexHor[1]], RowBegVer[IndexVer[1]], ColBegVer[IndexVer[1]], RowEndVer[IndexVer[1]], ColEndVer[IndexVer[1]], RowLR, ColLR, IsOverlapping)
• For visualization, we generate a contour that represents the fitted
• quadrilateral, i.e., the field border.
  gen_contour_polygon_xld (FieldBorder, [RowUL,RowUR,RowLR,RowLL,RowUL], [ColUL,ColUR,ColLR,ColLL,ColUL])
  dev_display (Image)
  dev_display (FieldBorder)
  disp_message (WindowHandle, ‘Fitted field borders (quadrilateral)’, ‘image’, -1, -1, ‘black’, ‘true’)
  disp_continue_message (WindowHandle, ‘black’, ‘true’)
  stop ()
• Now we can determine a projective transformation that will map our extracted
• quadrilateral to a rectangle of given dimensions. Note that we have to
• add 0.5 to the coordinates to make the extracted pixel positions fit the
• coordinate system that is used by perspective_trans_image.
  vector_to_proj_hom_mat2d ([RowUL,RowUR,RowLR,RowLL] + 0.5, [ColUL,ColUR,ColLR,ColLL] + 0.5, [160,160,340,340] + 0.5, [250,550,550,250] + 0.5, ‘normalized_dlt’, [], [], [], [], [], [], HomMat2D, Covariance)
• Note that we could also use the following call to determine the transformation:
• hom_vector_to_proj_hom_mat2d ([RowUL,RowUR,RowLR,RowLL] + 0.5, [ColUL,ColUR,ColLR,ColLL] + 0.5, [1,1,1,1], [160,160,340,340] + 0.5, [250,550,550,250] + 0.5, [1,1,1,1], ‘normalized_dlt’, HomMat2D)
• Now we can map the image and the quadrilateral.
  projective_trans_image_size (Image, TransImage, HomMat2D, ‘bilinear’, 800, 500, ‘false’)
  projective_trans_contour_xld (FieldBorder, FieldBorderTrans, HomMat2D)
  dev_set_window_extents (-1, -1, 800, 500)
  dev_display (TransImage)
  dev_display (FieldBorderTrans)
  disp_message (WindowHandle, ‘Rectified track and field:’, ‘image’, 20, 10, ‘black’, ‘true’)
  disp_message (WindowHandle, ‘virtual overhead view’, ‘image’, 45, 10, ‘black’, ‘true’)